www.HIMSSConference.org

#smartHIT

FEBRUARY 11, 2019

ORLANDO, FL

How AI Enabled a Community

Hospital to Tackle Clinical Variation

And Reduce Length-of-Stay

• Topics I will cover

• Why use AI for Managing Clinical Variation

• Topics I will cover

• What we at Flagler Hospital have done

• Topics I will cover

• How we operationalized it

• Topics I will cover

• Our Results

Why reducing Clinical Variation Is Important

• $750+ B wasted on unnecessary care

Why reducing Clinical Variation Is Important

• Medical mistakes are now the 3

leading cause of

death in the USA

Why reducing Clinical Variation Is Important

• Now the 3

leading cause of death in Canada

Why reducing Clinical Variation Is Important

• WHO reports 95,000 deaths in Europe due to

Medical Errors (15

2012 Institute of Medicine

67%

30%

Percent

Necessary

Unnecessary

Deaths

$750 B

Variation in Healthcare Expenditures 2011 AHA

Unexplained

Types of Variation

• Common Cause Variation

• Special Cause Variation

Types of Variation

• Common Cause Variation

• It is always present

• Special Cause Variation

Types of Variation

• Common Cause Variation

• It is inherent in the process

• Special Cause Variation

Types of Variation

• Common Cause Variation

• To change the results, change the process

• Special Cause Variation

Types of Variation

• Common Cause Variation

• Special Cause Variation

• It does not always happen in the process

Types of Variation

• Common Cause Variation

• Special Cause Variation

• Since it is so different, you would want to ask “why”

Types of Variation

• Common Cause Variation

• Special Cause Variation

• Eliminate or minimize the impact if negative

Types of Variation

• Common Cause Variation

• Special Cause Variation

• Not knowing the difference will create wasted time & effort

• Hospitals have tried for years to reduce clinical variation

• We gather the data we think is important

• Hospitals have tried for years to reduce clinical variation

• We analyze the data then try to implement our findings

• Hospitals have tried for years to reduce clinical variation

• By the time we act, the data may have changed significantly

• Over the past several years, three things have changed

• We now have massive computational power

• Over the past several years, three things have changed

• Our EMR has massive amounts of data

• Over the past several years, three things have changed

• AI systems like Ayasdi can now look at all the data and answer

questions we did not know to ask.

Methods of AI

• Supervised Learning

Methods of AI

• Unsupervised Learning

Methods of AI

• Ayasdi uses unsupervised learning and a branch of

mathematics called Topology

Methods of AI

• Euler, in 1736 solution to the 7 bridges of Konigsberg

Topological Map of our Pneumonia Pilot

Or Team

• What did we need to do?

• 2,500 lines of SQL code to extract the data

• What did we need to do?

• Upload to Ayasdi

• What did we need to do?

• Perform Semantic and Syntactic Validation

• What did we need to do?

• Generate the Treatment Groups

• What did we need to do?

• Select the “Goldilocks” Cohort

• What did we need to do?

• Ayasdi generates the CarePath GL Cohort

• What did we need to do?

• Begin monitoring our providers

61.1%

8.9%

14.8%

22.3%

9.8%

28.9%

20.0%

13.3%

38.8%

31.5%

24.1%

29.1%

36.6%

23.7%

25.7%

33.3%

31.0%

12.1%

18.5%

16.5%

24.4%

34.2%

25.7%

33.3%

2.3%

6.5%

4.6%

7.8%

7.3%

0.0% 0.0%

3.3%

0.0%

1.6%

1.9%

1.0%

0.0% 0.0%

2.9%

0.0%

10%

20%

30%

40%

50%

60%

70%

Cohort 216 Cohort 124 Cohort 108 Cohort 103 Cohort 41 Cohort 38 Cohort 35 Cohort 30

Comorbid Conditions

Pneumonia

Diabetes COPD CHF Hypotension Mortality

3.99

3.35

3.04

3.83

2.90

3.11

2.11

3.37

$3.17

$2.46

$2.32

$2.95

$2.42

$2.68

$2.01

$2.76

0.9%

0.0%

0.9%

0.0% 0.0% 0.0%

2.9%

0.0%

1.6%

1.9%

1.0%

0.0% 0.0%

2.9%

0.0%

0.5%

1.0%

1.5%

2.0%

2.5%

3.0%

3.5%

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

4.50

Cohort 216 Cohort 124 Cohort 108 Cohort 103 Cohort 41 Cohort 38 Cohort 35 Cohort 30

Cohort Evaluation

Pneumonia

LOS DVC/k Readmit Mortality

12.30

3.49

4.91

2.89

$13.05

$4.21

$5.10

$3.21

1.1%

0.0%

1.6%

0.0%

9.0%

34.8%

9.4%

24.6%

10%

15%

20%

25%

30%

35%

40%

0.00

2.00

4.00

6.00

8.00

10.00

12.00

14.00

Cohort 89 Cohort 69 Cohort 64 Cohort 57

Cohort Analysis

Septic Shock

LOS DVC/k Readmit Mortality

$35,310.79

$17,634.29

$49,316.00

$171,135.64

$(74,951.68)

$(100,000.00)

$(50,000.00)

$50,000.00

$100,000.00

$150,000.00

$200,000.00

Savings Booked to Date

Total Savings

$198,445.04

Pneumonia COPD CHF Septic Shock Sepsis w/o Shock

• What do you need to do

• Need the SQL skills to retrieve the data

• What do you need to do

• Bring physicians in early

• What do you need to do

• Recognized that it is an iterative process

• What do you need to do

• Work hard

• What do you need to do

• We can change the world!

www.HealthcareMachineLearningAI.com

#smartHIT

Michael C. Sanders, M.D.

Flagler Hospital

Michael.Sanders@FlaglerHospital.org